[boost] [Spirit] real_policies<> and 1.0#NAN

Subject: [boost] [Spirit] real_policies<> and 1.0#NAN
From: Pete Bartlett (pete_at_[hidden])
Date: 2016-06-07 17:00:51

Hi,

Boost.Spirit is not sure of itself when it comes to parsing representations of infs and nans such as “1.0#NAN”.

The docs [1] for the default real_policies<T> for parsing numbers say: “[…] with the additional detection of NaN and Inf as mandated by the C99 Standard and proposed for inclusion into the C++0x Standard: nan, nan(...), inf and infinity (the matching is case-insensitive). This corresponds to the following grammar:”

The first sentence suggests that 1.0#NAN etc will not be parsed. But the grammar itself then says:

<quote>
nan
   =   -lit("1.0#") >> no_case["nan"]
>> -('(' >> *(char_ - ')') >> ')')
   ;

inf
   =   no_case[lit("inf") >> -lit("inity")]
   ;
</quote>

suggesting that "#1.0NAN" is intended to parsed, but "#1.0INF" is not! 

Then the in-code docs [2] however suggest that both "1.0#NAN" and "1.0#INF" are intended to be parsed:

<quote>
       //  The second call allows to recognize representation formats starting
       //  with a 1.0 (such as 1.0#NAN or 1.0#INF etc.).
</quote>

However, the actual code fails to parse either of these these. This is because in [3] the presence of the "0" causes frac_digits to be non-zero and so the "traits::is_equal_to_one(..)" condition is never tested. In the case "1.#NAN" (without a zero) frac_digits is zero so the condition is tested and we enter the traits::parse_nan code, but now the text parsed to this sub-parser is "#NAN" and the parser does not cater for the initial "#".

Could the maintainers state which behaviour is intended...hopefully I may be able to supply a patch to docs/code/tests knowing that.

Regards,
Pete

[1] http://www.boost.org/doc/libs/1_61_0/libs/spirit/doc/html/spirit/qi/reference/numeric/real.html
[2] <boost/spirit/home/qi/numeric/real_policies.hpp>
[3]<boost/spirit/home/qi/numeric/detail/real_impl.hpp>